Cascaded MPC for Plantwide Optimization and Constraint Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional industrial process control and automation systems face challenges in implementing closed-loop plantwide optimization due to the lack of guaranteed solution consistency across multiple layers, leading to unreachables optimization benefits.
Innovation Solution
A cascaded model predictive control (MPC) approach is implemented, where a master MPC controller uses a planning model to perform plantwide optimization by sending optimization calls to slave MPC controllers and receiving proxy limit values to honor their constraints, enabling decentralized controls and centralized optimization in a single consistent system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If decentralized MPC solutions are implemented at lower levels, then operability and flexibility in dealing with process upsets and equipment failures are improved, but solution consistency across multiple layers deteriorates
Solution Approach 1:
The control system is segmented into multiple hierarchical layers with master MPC controllers at the plant level and slave MPC controllers at the unit level. Each layer operates semi-independently with its own optimization objectives and constraints, allowing decentralized adaptability while maintaining overall system consistency through hierarchical coordination.
Solution Approach 2:
Proxy limits are introduced as intermediary parameters that transmit constraint information from slave MPC controllers to master MPC controllers. These proxy limits act as mediators that enable the master controller to honor slave controller constraints without direct coupling, maintaining solution consistency across layers while preserving decentralized operability.
2Device complexity
If centralized planning optimization is implemented at higher levels, then a higher-level view that distills out unessential details is improved, but implementation as part of a closed-loop control system deteriorates
Solution Approach 1:
The centralized optimization function is segmented and distributed across master MPC controllers at different hierarchical levels. Each master controller handles plant-wide optimization for its respective section, breaking down the complex centralized problem into manageable segments that can be implemented in closed-loop while maintaining the benefits of high-level planning.
Solution Approach 2:
The optimization system transitions from static centralized planning to dynamic closed-loop control by implementing MPC at multiple levels. The master and slave MPC controllers continuously adjust their operations based on real-time process conditions, enabling centralized optimization to function as part of a dynamic closed-loop control system rather than as a static open-loop plan.
3Adaptability or versatility
If multiple hierarchical layers are implemented for different control functions, then specialized control and optimization functions are improved, but guaranteed solution consistency across layers deteriorates
Solution Approach 1:
Proxy limits serve as intermediary parameters that bridge the master and slave MPC controllers across hierarchical layers. These intermediaries carry constraint information from lower levels to higher levels, ensuring that optimization decisions at each layer are consistent with the operational constraints of other layers, thereby guaranteeing solution consistency across the entire hierarchical system.
Solution Approach 2:
The hierarchical MPC system implements feedback mechanisms where slave MPC controllers provide proxy limit information to master MPC controllers, which then adjust their optimization decisions accordingly. This feedback loop ensures that solution consistency is maintained across layers by continuously incorporating constraint information from lower levels into higher-level optimization decisions.
Data Source
AI summary
A method includes obtaining a planning model for an industrial facility at a master MPC controller and sending at least one optimization call from the master MPC controller to one or more slave MPC controllers. The method also includes receiving at least one proxy limit value from the slave MPC controller(s) in response to the at least one optimization call. The at least one proxy limit value identifies to what extent one or more process variables controlled by the slave MPC controller(s) are adjustable without violating any process variable constraints. In addition, the method includes performing plantwide optimization at the master MPC controller using the planning model and the at least one proxy limit value. The at least one proxy limit value allows the master MPC controller to honor the process variable constraints of the slave MPC controller(s) during the plantwide optimization.